Wireless Ad Hoc and Sensor Networks

Congestion Control in ATM Networks and the Internet 139general, it is very difficult to use offline training as the bursty MPEG datais statistically different from one source to the other.Novel weight-tuning methods are derived for the NN. A rigorous mathematicalanalysis is provided for the closed-loop system, defined in terms oferror in buffer occupancy, unlike in other available congestion control techniques.In fact, the proposed NN controller guaranteed performance as shownthrough the Lyapunov analysis. In addition, the performance in terms of CLRcan be reduced to arbitrarily small values by choosing the gains appropriately.Results are presented by evaluating the performance of the proposedcontroller during congestion, when both ON/OFF and bursty MPEGsources were transmitting data onto the network. Based on the results, itwas concluded that the proposed methodology generates a low CLR andoverall delay compared to the adaptive ARMAX, one layer, and the thresholdingmethods, and the NN controller ensured fairness during congestion.In addition, in most cases, one can design a switch with a smallbuffer due to the weight adaptation. In other words, simulation resultsshow that the performance of the proposed two-layer NN-based congestioncontroller varies insignificantly with an increase in buffer size. Thisis an important result, which can be used to design buffers for networkswitches. Finally, the learning-based methodology provides an efficientqueue management for ATM networks.Next, this chapter discussed a rate-based end-to-end congestion controlscheme for packet switched networks. The buffer dynamics were modeledas a nonlinear dynamical system, which is driven by the outgoing bandwidth,bottleneck queue level, and inflow rate. The bottleneck queue levelis estimated by packet-arrival rate at ingress/egress nodes. A rigorousmathematical analysis is provided for the closed-loop system, defined interms of error in queue length unlike in other available congestion controltechniques. In fact, the proposed controller guaranteed performance, asshown through the Lyapunov analysis. Simulation examples were givento justify the theoretical conclusions. Based on the results, it was concludedthat the proposed methodology performs better compared to theconventional New-Reno TCP method.ReferencesBae, J.J. and Suda, T., Survey of traffic control schemes and protocols in ATMnetworks, Proceedings of the IEEE, Vol. 79, No. 2, February 1991, pp. 170–189.Bansal, D. and Balakrishnan, H., Binomial congestion control algorithms, Proceedingsof the INFOCOM, Vol. 2, 2001, pp. 631–640.Benmohamed, L. and Meerkov, S.M., Feedback control of congestion in packetswitching networks: the case of a single congested node, IEEE/ACM Transactionon Networking, Vol. 1, No. 6, 693–707, 1993.